[ I'm teaching log-linear models at present, and looking at Agresti's
book. ]
Often when producing analyis of variance or analysis of deviance
tables it is important to get effects fitted in a specific order.
At present in S / R this is done as follows:
anova( glm( terms(formula, keep.order=T), ...) )
Can anyone think of a reason why keep.order should not become an
argument to "lm" and "glm", rather than forcing the use of "terms".
I.e., the above could be written:
anova( glm(formula, ..., keep.order=T) )
Having to resort to the use of "terms" seems like a complication my
punters could do without.